What unbound geo does not do.
Limits of the measurement
What a scan cannot tell you.
These are properties of the thing being measured, not features we have not built yet. None of them is on a roadmap, because none of them can be.
It cannot tell you why an assistant named somebody
We store what came back, not the model’s reasons for it, because the model does not have reliable access to those either. What we can show is which sources the answer cited and which brands it named, run after run. That is a pattern you can act on. It is not a causal explanation, and anybody selling you one is selling a guess with a confident voice.
It cannot promise to get you recommended
Nothing here is an optimisation service with a lever at the end of it. The product measures, dates and stores. What you do about the reading is your work, or your agency’s, and the honest position is that some of it will not move for reasons no measurement will surface.
It cannot report a change from a single run
Two snapshots differ reportably only when their confidence intervals do not overlap, and there is no interval under three runs. A first scan tells you where you stand. It does not tell you that you moved, and the product will not draw you an arrow that says you did.
It cannot see the conversation, only the answer
Reasoning models return a private draft alongside the reply, and that draft names brands the final answer then drops. We exclude it and record how many items we dropped. Counting it would report you as visible in a place no buyer ever reads, which is a larger number and a worse one.
It cannot cover a question nobody asks
A scan is only as good as the questions in it. If the set is written to make you look good, it will. The prompts are visible and editable for exactly that reason, and a question that never discriminates between brands is reported as uninformative rather than quietly counted as a win.
Rules we will not break
What the product will not do, even when it would flatter you.
Each of these makes the number smaller. That is the point: every one exists because we watched the alternative produce a figure nobody could defend in a room.
Generate anything with a model
Every figure comes from a real call to a real engine and the raw response is stored. Without a provider connected the feature stays inert and writes nothing — it does not fall back to an estimate, and there is no code path where a model is asked whether you look good.
Count a failed call as a zero
A call that failed is excluded from both sides of the ratio and the coverage is declared beside the number. Scoring failures as absences would make every outage look like a competitor’s good week.
Drop a competitor to improve your share
Every brand an answer names enters the denominator whether or not it is on your list. A share of voice that improves because you stopped tracking somebody is a corrupt number, and it is the easiest number in this category to produce by accident.
Match your brand name as a substring
We measured a brand whose name is also an ordinary word returning 2,420 mentions from an index while the brand’s own domain returned zero on the same query. Three orders of magnitude of noise, and the direction of the error always flatters the customer, which is why it survives in other tools.
When to walk away
Who should not buy this.
A page that concludes everybody should buy is one nobody believes. We would rather lose the deal we were going to lose anyway, earlier and with the reason stated.
You want a guarantee of placement
Nobody can sell you one. An assistant’s answer is not an ad slot, and anyone quoting you a position is describing something they do not control.
You need one number for a board slide and nothing under it
The product is built the other way round: every figure opens to the question, the engine, the model, the date and the full answer text. If the evidence is not the point for you, you are paying for the expensive half.
Your category is genuinely tiny
If assistants return the same two names to every phrasing a buyer could use and one of them is yours, you already know your position. Run the free scan, confirm it, and spend the money somewhere it changes a decision.
You want us to beat the established platforms at ordinary search
We do not. Rank tracking, keyword research and backlinks are included so the three surfaces sit in one report, not because the depth competes with a platform that has spent fifteen years on it. The comparison pages say so by name.
The questions, answered short
The ones that arrive before anybody signs up.
What can unbound geo not tell me?
Why an assistant named somebody. We store the answer and its cited sources, not the model’s reasons, because the model does not have reliable access to those either. We also cannot report a change from a single run: there is no confidence interval under three runs, so a first scan establishes a position rather than a trend.
Does it guarantee my business gets recommended by AI?
No, and nothing can. An assistant’s answer is not a purchasable slot. The product measures what is being said, dates it and stores the evidence. Acting on the reading is separate work, and some of it will not move.
How long does a free scan take?
Under ten minutes, with no account and no card. Five real questions go to a live engine with web search on and you get the full answer text back, not a summary of it.
What happens when an engine call fails?
It is excluded from both sides of the ratio and the coverage is declared next to the figure. A failed call is not scored as an absence, because that would turn every provider outage into a competitor’s good week.
Why is my score sometimes lower than another tool says?
Usually one of three rules. We match whole words rather than substrings, we exclude the reasoning trace that names brands the answer drops, and we count every brand an answer names in the denominator including ones you never listed. All three make the number smaller and all three make it defensible.
Who should not buy this?
Anyone who wants guaranteed placement, anyone who wants a single number with no evidence under it, and anyone in a category so small that they already know the two names an assistant returns. Also anyone expecting us to out-depth the established SEO platforms at ordinary search — we do not, and the comparison pages say so.
Is the site audit included on the free plan?
No. It is the one capability that touches somebody else’s server on request, so it does not run on an account nobody has verified. Every other capability is available from the entry plan upward; the ladder is rows and refresh rate, not access.
What did your own first full run actually cost?
15 questions across 4 engines for $4.01, which stored 51 answers and 460 cited URLs across 190 domains. The ledger reconciled exactly against the provider. We publish it because a vendor who will not tell you what a call costs is a vendor with a margin to protect.
Figures on this page are from stored runs, measured 30 July 2026. Nothing here is generated by a model guessing.
Still the right tool for most people in this position.
Knowing what something cannot do is how you find out whether it does the thing you needed. Run the free scan and decide against your own category rather than against a page.